ultralytics 8.3.209__tar.gz → 8.3.210__tar.gz

{ultralytics-8.3.209 → ultralytics-8.3.210}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.209
+Version: 8.3.210
 Summary: Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>

{ultralytics-8.3.209 → ultralytics-8.3.210}/ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.209"
+__version__ = "8.3.210"
 
 import importlib
 import os

{ultralytics-8.3.209 → ultralytics-8.3.210}/ultralytics/engine/trainer.py

@@ -420,9 +420,10 @@ class BaseTrainer:
                     self.loss = loss.sum()
                     if RANK != -1:
                         self.loss *= self.world_size
-                    self.tloss = (
-                        (self.tloss * i + self.loss_items) / (i + 1) if self.tloss is not None else self.loss_items
-                    )
+                    if not self.loss.isfinite():
+                        LOGGER.warning("Non-finite forward pass, skipping batch...")
+                        continue
+                    self.tloss = self.loss_items if self.tloss is None else (self.tloss * i + self.loss_items) / (i + 1)
 
                 # Backward
                 self.scaler.scale(self.loss).backward()

{ultralytics-8.3.209 → ultralytics-8.3.210}/ultralytics/nn/tasks.py

@@ -743,6 +743,22 @@ class RTDETRDetectionModel(DetectionModel):
         """
         super().__init__(cfg=cfg, ch=ch, nc=nc, verbose=verbose)
 
+    def _apply(self, fn):
+        """
+        Apply a function to all tensors in the model that are not parameters or registered buffers.
+
+        Args:
+            fn (function): The function to apply to the model.
+
+        Returns:
+            (RTDETRDetectionModel): An updated BaseModel object.
+        """
+        self = super()._apply(fn)
+        m = self.model[-1]
+        m.anchors = fn(m.anchors)
+        m.valid_mask = fn(m.valid_mask)
+        return self
+
     def init_criterion(self):
         """Initialize the loss criterion for the RTDETRDetectionModel."""
         from ultralytics.models.utils.loss import RTDETRDetectionLoss

{ultralytics-8.3.209 → ultralytics-8.3.210}/ultralytics/utils/benchmarks.py

@@ -583,6 +583,11 @@ class ProfileModels:
         run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=3)  # sigma clipping
         return np.mean(run_times), np.std(run_times)
 
+    @staticmethod
+    def check_dynamic(tensor_shape):
+        """Check whether the tensor shape in the ONNX model is dynamic."""
+        return not all(isinstance(dim, int) and dim >= 0 for dim in tensor_shape)
+
     def profile_onnx_model(self, onnx_file: str, eps: float = 1e-3):
         """
         Profile an ONNX model, measuring average inference time and standard deviation across multiple runs.
@@ -604,27 +609,36 @@ class ProfileModels:
         sess_options.intra_op_num_threads = 8  # Limit the number of threads
         sess = ort.InferenceSession(onnx_file, sess_options, providers=["CPUExecutionProvider"])
 
-        input_tensor = sess.get_inputs()[0]
-        input_type = input_tensor.type
-        dynamic = not all(isinstance(dim, int) and dim >= 0 for dim in input_tensor.shape)  # dynamic input shape
-        input_shape = (1, 3, self.imgsz, self.imgsz) if dynamic else input_tensor.shape
-
-        # Mapping ONNX datatype to numpy datatype
-        if "float16" in input_type:
-            input_dtype = np.float16
-        elif "float" in input_type:
-            input_dtype = np.float32
-        elif "double" in input_type:
-            input_dtype = np.float64
-        elif "int64" in input_type:
-            input_dtype = np.int64
-        elif "int32" in input_type:
-            input_dtype = np.int32
-        else:
-            raise ValueError(f"Unsupported ONNX datatype {input_type}")
+        input_data_dict = dict()
+        for input_tensor in sess.get_inputs():
+            input_type = input_tensor.type
+            if self.check_dynamic(input_tensor.shape):
+                if len(input_tensor.shape) != 4 and self.check_dynamic(input_tensor.shape[1:]):
+                    raise ValueError(f"Unsupported dynamic shape {input_tensor.shape} of {input_tensor.name}")
+                input_shape = (
+                    (1, 3, self.imgsz, self.imgsz) if len(input_tensor.shape) == 4 else (1, *input_tensor.shape[1:])
+                )
+            else:
+                input_shape = input_tensor.shape
+
+            # Mapping ONNX datatype to numpy datatype
+            if "float16" in input_type:
+                input_dtype = np.float16
+            elif "float" in input_type:
+                input_dtype = np.float32
+            elif "double" in input_type:
+                input_dtype = np.float64
+            elif "int64" in input_type:
+                input_dtype = np.int64
+            elif "int32" in input_type:
+                input_dtype = np.int32
+            else:
+                raise ValueError(f"Unsupported ONNX datatype {input_type}")
+
+            input_data = np.random.rand(*input_shape).astype(input_dtype)
+            input_name = input_tensor.name
+            input_data_dict.update({input_name: input_data})
 
-        input_data = np.random.rand(*input_shape).astype(input_dtype)
-        input_name = input_tensor.name
         output_name = sess.get_outputs()[0].name
 
         # Warmup runs
@@ -632,7 +646,7 @@ class ProfileModels:
         for _ in range(3):
             start_time = time.time()
             for _ in range(self.num_warmup_runs):
-                sess.run([output_name], {input_name: input_data})
+                sess.run([output_name], input_data_dict)
             elapsed = time.time() - start_time
 
         # Compute number of runs as higher of min_time or num_timed_runs
@@ -642,7 +656,7 @@ class ProfileModels:
         run_times = []
         for _ in TQDM(range(num_runs), desc=onnx_file):
             start_time = time.time()
-            sess.run([output_name], {input_name: input_data})
+            sess.run([output_name], input_data_dict)
             run_times.append((time.time() - start_time) * 1000)  # Convert to milliseconds
 
         run_times = self.iterative_sigma_clipping(np.array(run_times), sigma=2, max_iters=5)  # sigma clipping

{ultralytics-8.3.209 → ultralytics-8.3.210}/ultralytics.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ultralytics
-Version: 8.3.209
+Version: 8.3.210
 Summary: Ultralytics YOLO 🚀 for SOTA object detection, multi-object tracking, instance segmentation, pose estimation and image classification.
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>